Limit captured hook output from prek run -v to failed hooks

[pygarap]

Summary

Without any verbosity level, prek run already shows captured hook output only for failed hooks. However, a non-verbose run does not include the additional hook metadata provided by -v / --verbose.

The request is for verbosity level 1 to add that metadata while preserving the default failed-hook-only captured output behavior.

Currently, prek run -v prints nonempty captured output for passed hooks as well as failed hooks. Although --verbose is a counting flag, prek run receives only cli.globals.verbose > 0, so -v, -vv, and -vvv have the same captured hook output behavior.

Keep level 1 focused on hook metadata and timing, and reserve captured output from non-failed hooks for level 2 and above.

Requested behavior

Invocation	Hook metadata	Captured hook output
No verbosity	No verbose metadata	Failed hooks only
-v or --verbose	Hook ID, alias, description, and duration	Failed hooks only
-vv, repeated --verbose, or higher	Hook ID, alias, description, and duration	Any hook with nonempty output

At level 1, -v / --verbose should continue adding the existing metadata for every hook result other than a (no files to check) Skipped result:

1. Hook ID

   - hook id: <id>
   

2. Hook alias, when nonempty and different from the hook ID

   - hook alias: <alias>
   

3. Hook description, when configured

   - description: <first line>
   

4. Execution duration

   - duration: <duration>s
   

Captured hook output at level 1 should behave exactly like a normal run without verbosity: show it for failed hooks, but suppress it for Passed, Dry Run, and other non-failed results.

Additional requirements:

• Preserve verbose: true as an explicit per-hook override.
• Preserve the existing log_file behavior.
• Do not change internal debug or trace logging at level 2 and above.

Example

repos:
  - repo: local
    hooks:
      - id: passing-output
        name: Passing output
        language: system
        entry: python3 -c 'print("captured pass output")'
        always_run: true
        pass_filenames: false
      - id: failing-output
        name: Failing output
        language: system
        entry: python3 -c 'print("captured failure output"); raise SystemExit(1)'
        always_run: true
        pass_filenames: false

Usage

At level 1, the verbose metadata should be shown for the hooks, but only captured failure output should be printed:

$ prek run --all-files -v

At level 2, the same metadata should be shown and both captured outputs should be printed:

$ prek run --all-files -vv

Implementation note

Pass the counted verbosity level into the prek run result renderer instead of reducing it to a boolean.

[j178]

Without -v, this is already the default behavior, prek only shows output from failed hooks.

[pygarap]

@j178

Yes, I understand that without any verbosity level, prek already shows captured output only for failed hooks.

The distinction I am asking for is that running without -v also means you do not get the additional hook metadata provided by -v / --verbose.

What I would like is for -v / --verbose to provide that metadata:

1. Hook ID

   - hook id: <id>
   

2. Hook alias, when nonempty and different from the hook ID

   - hook alias: <alias>
   

3. Hook description, when configured

   - description: <first line>
   

4. Execution duration

   - duration: <duration>s
   

while still showing captured hook output only for failed hooks, just like a normal run without verbosity.

Then -vv and above could keep the current behavior of also showing captured output for non-failed hooks.

[j178]

For failed hooks, the hook ID, alias, and description are already shown. The only information that's limited to verbose mode is the duration, and I think it makes sense to keep it that way.

So unless I'm missing something, I think the behavior you're asking for is already supported?

[pygarap]

@j178 I think the misunderstanding is specifically about captured hook output.

I'm not asking for any changes to:

• No verbosity

• -vv, repeated --verbose, or higher verbosity levels

• The existing level-1 verbose metadata:

  • hook ID
  • hook alias
  • hook description
  • execution duration

At -vv, repeated --verbose, or higher verbosity levels, captured output would continue to be printed for non-failed hooks exactly as it is today.

The only change I'm requesting is for captured hook output when using exactly -v or --verbose.

For captured hook output only, -v should behave the same as a normal run with no verbosity:

• Show captured output for failed hooks
• Suppress captured output for Passed, Dry Run, and other non-failed results

So -v would still add the existing verbose metadata and execution duration. The only part that would follow no-verbosity behavior is captured hook output: it would not be additionally printed for successful or otherwise non-failed hooks.

[j178]

I understand the distinction, but I think non-verbose mode is the better fit for this use case. It already provides the failed-hook-only output behavior you want, whereas -v is intended to expose additional diagnostic information. I don’t think we need a separate captured-output policy for -v and -vv.

[pygarap]

@j178 What I’m really looking for is a way to print the basic metadata for every hook, including non-failed hooks:

• hook ID
• hook alias
• hook description
• execution duration

but without printing the captured hook output, which can be very long and add a lot of noise.

Non-verbose mode does not provide that metadata for non-failed hooks, while -v currently also prints their captured output.

So currently there is no verbosity level that provides the basic metadata for all hooks without also printing captured output for non-failed hooks. That is the gap I was trying to address.

If changing the meaning of -v is undesirable, perhaps an opt-in CLI flag such as --hide-captured-output could provide exactly this behavior while keeping the existing verbosity semantics unchanged, what do you think? Thanks 🙏 See: #2646